The RTX 4090 graphics card from Nvidia makes some bold claims. Positioned as the flagship of Nvidia’s brand new Ada Lovelace architecture, the RTX 4090 aims to be the fastest GPU on the market, whether you’re looking for just blazing fast framerates in heavy game workloads, or are a creative making some of the world’s best digital art. But do the claims match up with real-world performance? We take a look in our RTX 4090 review.
Specifications
- Graphics Processing Clusters: 11
- Texture Processing Clusters: 64
- Streaming Processors: 128
- CUDA Cores: 16384
- Tensor Cores: 512 (4th Generation)
- RT Cores: 128 (3rd Generation)
- Texture Unites: 512
- ROPs: 176
- Boost Clock: 2520 MHz
- Memory Clock: 10501 MHz
- Memory Data Rate: 21Gbps
- L2 Cache Size: 73728 KB
- VRAM: 24 GB GDDR6X
- Memory Interface: 384-bit
- Memory Bandwidth: 1008 GB/s
- Texture Rate (Bilinear) 1290.2 GigaTexels/second
- Fabrication Process: TSMC 4 nm Nvidia Custom Process
- Transistor Count: 76.3 billion
- I/O: 3 x DisplayPort 1.4a; 1 x HDMI 2.1a
- Power connector: 1 x 16-pin (comes with dongle with 4 x 8-pins)
- Minimum Recommended PSU: 850 Watts
- Total Graphics Power: 450 Watts
- Maximum GPU Temperature: 90C
- MSRP: $1599
Design
The RTX 40-series cards retain the look of last generation’s Ampere 30 series cards featuring a dual-axial blow-through design and minimalistic accents. The accents grace the front and sides of the card, while the twin fans on either side allow the powerful card to keep quiet and cool under load.
The fans on the RTX 4090 are a mite bigger than those on the Ampere RTX 3090, and while the board design looks pretty much identical to its Ampere brethren, small changes were made to ensure good airflow and performance under pressure. The larger fans now use fluid dynamic bearings and counter-rotating fans in order to keep the GPU quiet while under load to maintain cool temperatures. This, coupled with slightly more distance between the heatsink fins on the 4090, should mean airflow will be consistent across the card.
The same möbius strip that enveloped the 30-series Founder’s Edition GPUs is back with the RTX 4090 Founder’s Edition, and it looks as stunning as ever. While I love the aggressive gamer aesthetic of yesteryear still seen on some GPUs and accessories, the minimalistic approach by Nvidia has grown on me over the years. The small non-RGB lighting accents in the center of the GPU as well as the GeForce RTX logo on the side also adds to the minimalism.
The RTX 4090 Founder’s Edition comes equipped with three DisplayPort 1.4a ports and 1 HDMI 2.1a port allowing for high refresh rates at high resolutions with HDR. Additionally, the RTX 4090 has a 16-pin power connection port and comes with an octopus dongle which splits into four 8-pin connectors for standard power supplies.
It’s been widely noted that the RTX 4090 Founder’s Edition is a big card. In fact, it might necessitate a new case for many prospective consumers thanks to how long and wide the GPU sits. Indeed, my Lian Li O11 Dynamic case was long enough to handle the card, but with the octopus cable attached to the side of the GPU, I couldn’t even put the tempered glass side panel back on without possibly bending the cable in a weak point. This is a big, BIG card, comically so in fact. Each time I look at my case, I can’t help but chuckle at just how large the 4090 FE truly is. This is a three-slot card and, as a result, it’s one of the only GPUs which hasn’t sagged in my motherboard thanks to it being a true 3-slot card as compared to others that take up the space of three slots, but only require a two-slot connection to the case.
All this size and power means that you’ll need a good power supply. Thankfully, the Ada Lovelace architecture aims to provide more power than the Ampere-based 3090 Ti at the same Total Graphics Power (TGP) of 450 Watts, which we’ll get to later on in this review.
The Ada Lovelace architecture (named after the engineer of the same name) is being manufactured on TSMC’s new custom 4N process. As a result Nvidia has been able to fit 76.3 billion transistors into the chip. This is in addition to the increase in CUDA cores to 16384 versus the RTX 3090 Ti’s 10496. Additionally, the RTX 4090 includes third-generation RT Cores and fourth-generation Tensor Cores. Nvidia boasts “2x faster” ray-triangle intersection testing in Ada over Ampere.
The RTX 4090 FE has a boost clock up to 2520 MHz, up from the 3090 Ti’s 1860Mhz, and sports a full suite of 24 GB GDDR6X memory with 1TB/s bandwidth. The 4090 FE also has 12x the L2 cache over the 3090 Ti, sitting at a whopping 72 MB versus last generation’s 6.1 MB.
But all this power on paper wouldn’t be much if Nvidia wasn’t innovating on the technology driving it as well. So what new features does the Ada Lovelace card bring to the table?
Features Set
The Nvidia Ada Lovelace architecture brings with it multiple technological innovations to help power the GPU in any workload. This includes a new streaming multiprocessor (SM) that aims to give more performance and power efficiency to the GPU.
On top of these architectural breakthroughs, a new Shader Execution Reordering engine, or SER, aims to improve ray tracing by allowing the GPU to reschedule workloads on the fly, allowing for more efficient rendering.
DLSS 3
The big feature many gamers will wonder about is the next evolution of Nvidia’s groundbreaking Deep Learning Super Sampling, or DLSS. DLSS 3 aims to be the next step in neural graphics, adding to the already great results of simply reconstructing images of games using AI and machine learning, but creating whole new frames in the process.
DLSS 3 takes the image being produced using DLSS Super Resolution (DLSS 2) and uses the data from each frame, such as motion vectors, along with its AI algorithms to create an entirely new frame. It’s important to understand this frame is not created by the game’s engine. It is created entirely by DLSS 3. Additionally, these are not “fake” frames. These are real frames that you can objectively measure and perceive. This is done by utilizing the Ada Lovelace architecture’s Optical Flow Accelerator, which provides the neural network with the direction and velocity of pixels in each frame.
Since the DLSS 3 frame generation doesn’t interact with the graphics pipeline, it benefits not only the GPU itself but also the CPU. As a result, titles that are more CPU-bound, such as Microsoft’s Flight Simulator which uses a ton of physics calculations and streaming of real-world map data to fill out its world, can benefit from framerates that even the CPU could not compute otherwise.
However, introducing these extra generated frames could, in effect, introduce latency into the process for the player. This is where Nvidia is leveraging its existing Reflex technology to help counteract this added latency and make DLSS Frame Generated titles feel just as responsive as if these were traditionally generated frames from the graphics pipeline.
Nvidia Studio, RTX Remix, and Encoder Benefits
Nvidia Studio is also seeing benefits from the Ada Lovelace architecture on the RTX 4090 FE. This includes improvements to performance in workloads such as 3D rendering, video editing, and live streaming, the latter able to take advantage of Nvidia’s inclusion of open-source AV1 encoders. These encoders are more efficient than the existing H.264 encoders found on previous Nvidia GPUs, and aim to increase the quality of streams and broadcasts, such as livestreams in OBS or when using Nvidia Broadcast.
Additionally, creative users of the 40-series are going to benefit from DLSS 3 helping to empower workflows, making it easier and more efficient to edit and render in real-time when creating 3D art thanks to DLSS 3 and the 3rd Gen RT cores in the Lovelace architecture.
RTX Remix will also empower modders to take their creativity to new heights, allowing them to leverage the RTX pipeline and use AI to upscale textures in older games when creating mods.
RTX 4090 Synthetics, Benchmarks, and Thermal Performance
How does this all stack up in practice, though? We put the RTX 4090 through its paces in a series of gaming and synthetic benchmarks, comparing it with an array of 30-series GPUs as well as AMD’s 6000 series of cards. A note about our GPU suite: due to the nature of our site and our hardware testing, I currently do not have access to either an Nvidia RTX 3090 Ti nor the AMD Rx 6900 XT. However, we do have comparisons with many of the top-tier RTX and AMD cards to include here to glean gen-on-gen improvement, as well as comparison with the competition. Chris, our hardware editor, will be able to provide direct comparisons to the RTX 3090 Ti in future reviews on his bench.
Test Bench:
- CPU: Intel i7-10700K @ 3.8GHz (Boost Clock up to 4.8 GHz)
- Cooling: Corsair 100i 240mm Liquid Cooler
- Motherboard: Gigabyte Aorus Z490 Ultra
- RAM: Corsair Vengeance 32GB @ 3200MHz
- Storage: Intel 760p 2TB M.2 NVMe, ADATA Falcon 1TB M.2 NVMe
- PSU: Gigabyte 1200 W Platinum PSU
- Case: Lian Li 011 Dynamic
Let’s get this out of the way: The RTX 4090 FE is a 4K card. If you’re looking to buy this GPU for gaming, you’re looking at unlocking high refresh rate, raytraced 4K titles. However, we also opted to test at 1440p as it’s still a widely used resolution by many players out there and we wanted to show the numbers for each resolution when we could.
In our test bench, we compared the RTX 4090 FE against the higher-end 30-series cards from Nvidia, including:
- RTX 3080 Ti Founder’s Edition
- Gigabyte RTX 3080 Gaming OC 10GB
- EVGA RTX 3090 FTW3 Ultra
AMD cards included:
- AMD Rx 6800 XT Reference Card
- AMD Rx 6800 Reference Reference Card
- Sapphire NITRO+ Rx 6750 XT (replacing the 6700 XT in our bench as it’s a mid-cycle refresh).
We ran through a number of titles from various game engines, specifically looking at a mix of DX11 and DX12 titles. We also ran a suite of 3DMark’s Synthetic benchmarks showing relative performance in DX11, DX12, and Raytracing applications.
Where we could as well, we used in-game benchmark tools to ensure that each pass was as identical as the last. Where we couldn't, such as Control, A Plague Tale: Requiem , and Microsoft Flight Simulator 2020, we used a predefined path and replicated this each time. Additionally, it should be noted that each test that used AMD's FidelityFX Super Resolution and Nvidia's Deep Learning Super Sampling was done at the Performance preset.
We also ensured our 10700K was running with ReSizeable BAR enabled to get the most out of the performance when we could.
RTX 4090 Founder’s Edition Synthetic Benchmarks
In both FireStrike Ultra and Timespy Extreme, both 4K benchmarks, the RTX 4090 is by far the winner compared to the other cards in the bench. In Firestrike, we see a 68% increase in the overall score compared to the EVGA RTX 3090 FTW3, while in TimeSpy we see a 49% increase. Those jumps are mostly replicated when comparing it to the AMD Rx 6800 XT, the most powerful AMD card in our testing suite.
Synthetic benchmarks only tell part of the story, how does it stack up in real-world gaming?
RTX 4090 Founder’s Edition Raster Gaming Benchmarks
Across the board, especially at 4K, the RTX 4090 Founder’s Edition comes out on top in every single test, sometimes by a wide margin. 1440p we start to notice some potential CPU bottlenecking, most noticeably in Ubisoft’s Watch Dogs Legion, where the 1440p numbers were the closest between the 40-series and its 30-series brethren, suggesting the GPU was not fully utilized.
Impressively, titles such as Total War: Warhammer 3 show a massive improvement, especially at 4K where we see a 64% increase in average framerate compared to the EVGA RTX 3090 FTW3, and an ever wider margin compared to the 6800 XT with 136% improved. Intense titles like Horizon Zero Dawn, sees improvements over the 3090 at 4K to the tune of 128fps average compared to the 3090’s 98fps, a 31% increase.
Cyberpunk 2077, is one of the most impressive results we witnessed. As the most intense title in our bench, the gen-on-gen uplift in performance from Ampere to Ada Lovelace is incredible. Without any ray tracing enabled, we see a 52% increase over the 3090 at 4K, moving from 48fps average on the 3090 to 73fps on the 4090 FE.
Speaking of Raytracing, though…
RTX 4090 Founder’s Edition Raytracing Benchmarks
The 3rd Generation RT cores are doing work in Port Royal, the synthetic benchmark from 3DMark. We see an 85% increase in the score between the 4090 and 3090 in our bench, a massive increase over the competent 30 series.
First, a quick note on DLSS 2. With the introduction of DLSS 3 comes the “rebranding” of DLSS 2 as Super Resolution. In short, DLSS 2 will simply be a subset of DLSS 3 and will be called Super Resolution. All DLSS 3 games will support DLSS 2 (Super Resolution) if you don’t have a 40-series card.
With our ray tracing benchmarks, we see once again the 4090 come out on top across the board, though we do notice some interesting results with Watch Dogs Legion. As the RT title that is doing the “least” across the board (ray traced reflections only), we noticed in our testing that, especially with DLSS enabled, the GPU wasn’t being pushed to its limits compared to the CPU. This could be down again to a CPU bottleneck limiting performance (which the Watch Dogs Legion benchmark report also states as an impact).
However, the uplift without DLSS at 4K is pretty interesting: a 66% increase at 4K with ray tracing enabled versus the 3090.
Control, Remedy’s intense third-person shooter from a few years back, gives the suite a workout with it enabling multiple different ray tracing techniques, from lighting, shadows, and even debris reflections. Here too we see a massive uplift gen-on-gen across the board (though DLSS 1440p numbers are again, likely being CPU bottlenecked), with a 70% increase at 4K DLSS versus the 3090. When looking at the RT numbers with no upscaling enabled, we’re seeing a 221% increase in performance compared to the 6800 XT (61fps versus AMD’s 19fps average).
Cyberpunk 2077 is by far the most impressive here, though, as we start to see the benefits of DLSS 3 in action as well (more on that in a moment). When looking at just the 4K numbers, the RTX 4090 Founder’s Edition beats out every single card by a wide margin, regardless of the test setup. RT with DLSS disabled is almost a doubling of the 3090 result. With DLSS Super Resolution enabled (and DLSS Frame Generation disabled), we’re seeing a 57% increase in average framerates.
RTX 4090 Founder’s Edition DLSS 3 Tests
The RTX 4090 can also leverage frame generation, which further improves the margins of performance across the board in games where it’s supported. As a result, those same Cyberpunk 2077 gains are pushed out even further.
Even when just comparing the RTX 4090 to itself, we’re seeing a 259% increase in performance from 4K RT enabled and DLSS 3. That alone is an insane number, with the gen-on-gen uplift here still showing a 146% increase in performance (DLSS Super Resolution enabled on the 3090 vs the 4090’s DLSS 3).
In other titles we were able to test DLSS 3’s Frame Generation on, we see similar results play out. In F1 2022, full ray tracing enabled across the board, we see an 82% increase with DLSS 3 on the RTX 4090 at 4K versus the DLSS 2 test on the RTX 3090. Remember, DLSS 2 is effectively called DLSS Super Resolution moving forward.
Flight Simulator also bears this out. While not using ray tracing, it’s a great example of an incredibly CPU-bound title benefitting from DLSS 3 and how it can navigate the pipeline to provide higher than otherwise capable framerates.
Chinese MMO Justice gave us a real-time path tracing demo, showcasing what future real-time game raytracing can do on the 4090. Interestingly enough, it’s also using DLSS 3, and when compared with itself, the RTX 4090 sees a 4x increase when DLSS is disabled versus when it’s enabled with frame generation enabled. Funnily enough, I could not get the test to even work on the 30-series cards when DLSS was disabled, showing just how insanely heavy a workload this demo truly is.
As an aside, can we please have this MMO brought to the West? It’s so dang pretty, I want to explore more of the world.
Asobo’s upcoming A Plague Tale: Requiem also gets the DLSS 3 treatment, and while it doesn’t have any raytracing enhancements, the RPG benefits from frame generation as well, seeing a massive jump from 70fps to 166fps on the 4090 at 4K when toggling DLSS 3 on. Compared to the 3090 it’s a whopping 107% increase, while the RTX 3080 Ti, no slouch on its own, it’s a jump from 76fps to the 4090’s 166 – 118% increase for the 4090.
RTX 4090 Founder’s Edition Thermal And Power Performance
So for a card that has this much power, you’d assume it runs rather hot, right?
Well, the dual-axial blower design was carried over for a reason: the thermals on this thing are insanely cool for what we’re throwing at it over and over. Throughout my testing it never peaked over 70 degrees Celsius, even in long sessions at 4K Psycho Ray Tracing in Cyberpunk 2077. Now, keep in mind I was unable to get the side panel of my case back on the system, but it’s still a rather cool number, considering.
There has also been much made about the power draw of the RTX 4090 FE, especially as rumors ahead of Nvidia’s announcement last month made it seem as though this card would be rated at a TGP of 600 Watts. Crucially, the 4N process at TSMC has been able to pack more performance into this board at the same 450W power draw of the 3090 Ti. It’s still recommended to have a large power supply, especially if you plan on overclocking this GPU. But in my testing, I never saw a wattage over 410W, as reported via Nvidia’s API using FrameView, which is rather accurate in the long run.
That is still a lot of power, but considering what we’re seeing performance-wise, for Nvidia to do this at the same rating as the previous gen’s best card is rather impressive on the whole.
RTX 4090 Founder’s Edition Final Thoughts
So what does this all mean? The RTX 4090 Founder’s Edition is, without a doubt, the fastest card on the market when it hits shelves on October 12th. At every turn, it beat out the competition and shows a true generational leap over the Ampere architecture.
However, what truly feels most transformative is DLSS 3. At first, I was skeptical watching Nvidia’s presentation, wondering just how good DLSS 3 would feel. Sure, it can create frames, but would there be any semblance of latency while playing compared to the normal way of rendering titles?
But by leveraging Nvidia Reflex, in my hours of in-game testing, I honestly cannot tell a difference between how it feels with DLSS frame generation on or off. It’s a truly transformative leap in an already incredible technology, and I’m all for every developer adding this to their games in the months and years to come.
To play Cyberpunk 2077 with Psycho RT at over 130fps on the 40-series, when a week ago I was barely getting by in the low 50s (still good, but c’mon) on my RTX 3090, I can’t see myself going back. We’ve said it before and we’ll say it again: DLSS should be implemented in every single PC game. And when DLSS 3 launches, it will do so with over 35 games and engines supporting the feature.
Even without DLSS 3, especially since not every game is going to include it off the bat, the improvements in just pure rasterization and RT over the 30 series are incredible. And we expect that, right? There is always a gen-on-gen improvement when we see a new architecture release, whether it’s AMD or Nvidia.
With all of the other additions for creators, from the dual AV1 encoders to the inclusion of RTX Remix for modders, the 40-series of GPUs looks to bring a new generation of visuals and power to users everywhere.
But what about the cost?
$1,599 is a lot of money, especially in an economy dealing with high inflation across. Couple this with the fact that previous generation GPUs are dropping in price as the crypto market crashes and more stock just becomes available, it begs the question of whether this card is worth the money.
But it also raises a related question: who is this card for?
For creatives looking for the best workstation card to create intense 3D worlds, or the video content creator looking for the best quality livestream, this card could be for you. Gamers out there who demand the best, regardless of price, the 4090 is certainly that right now.
It’s also important to note that the card Nvidia is making the most comparisons to, the RTX 3090 Ti, retailed for $1,999 when it was released. We also don’t know what AMD’s upcoming Radeon stack is going to shake out with but given how its priced other GPUs in the recent past (I’m looking at you, Rx 6750 XT), the sticker shock of the RTX 4090 might be lessened some if AMD also sees the need to price their flagship cards higher than before.
There’s also a subset of fans who have the means to buy this card, no matter the price, and Nvidia on some level knows that. Many gamers are going to be looking ahead to the RTX 4080 16GB and 12GB variant launches later this year instead of the RTX 4090, which will still benefit from the Ada Lovelace improvements, DLSS 3, and more.
This isn’t a card that every gamer will be looking at buying, and that’s okay.
As it stands, the massive uplift in performance across the board for the RTX 4090 Founder’s Edition is enough for me to recommend to someone looking to buy this card, especially if you were hanging onto an old RTX 2080 Ti, or even an RTX 3090 (I seriously just bought my 3090 two months ago and I don’t see it going back into my PC any time soon). It's a considerable upgrade gen-on-gen versus Ampere that brings legitimately transformative technology to gamers and creators in DLSS 3 and much more.
Just be ready to buy a new case when you do.
Full Disclosure: The product discussed was provided by PR for the purposes of this review.
